Molecular dynamics driving phenotypic divergence among KRAS mutants in pancreatic tumorigenesis

SUMMARY

Inflammation in the pancreas drives acinar-to-ductal metaplasia (ADM), a progenitor-like state that can be hijacked by mutant Kras in the formation of pancreatic cancer (PDAC). How these cell fate decisions vary according to KRAS mutation remains poorly understood. To define mutation-specific lineage reversion and tumor initiation, we implement Ptf1a-TdTomato mice and multiple KRAS mutants across several genetic, pharmacologic, and inflammatory perturbations in vivo. Whereas KRAS^G12D co-opts injury to enable lineage reversion, enhancer reprogramming, and tumor initiation, KRAS^G12R/V cannot sustain dedifferentiated and neoplastic transcriptional and epigenetic programs. Specifically, KRAS^G12R/V mutants fail to invoke robust EGFR, AKT, and RAC1/VAV1 signaling and to license Pou2f3 and Vav1 in chromatin, such that only constitutive AKT activation is sufficient to rescue the tumorigenic potential of KRAS^G12R in vivo. As the marked heterogeneity among KRAS variants begins early in tumorigenesis, these data are crucial to deciphering mutation-specific oncogenic trajectories and directing the implementation of KRAS-directed therapeutics.

Graphical Abstract

eTOC

Grimont et al. demonstrate that enhancer reprogramming and EGFR and RAC1 signaling diverge between mutant KRAS alleles in the pancreatic epithelium. This leads to a differential ability to usurp injury-induced chromatin rewiring in tumorigenesis, establishing the presence of distinct oncogenic trajectories tied to the precise driver mutation in pancreatic cancer.

INTRODUCTION

Living systems exist in an apparent contradiction – they ensure stability through a profound capacity for change. Indeed, in the pancreas, the cellular plasticity of the acinar compartment – namely to undergo acinar-ductal metaplasia (ADM) and regeneration in the context of inflammatory injury – is essential to tissue homeostasis.^1–3 Multiple reports have suggested that ADM is not a bona fide transdifferentiation event, but one that features re-expression of pancreatic progenitor markers, including Nestin, KLF4, and KLF5^1,4,5 that can support tumorigenesis^5–7. However, the extent to which lineage reversion – apart from expression of these selected markers – is a conserved feature of inflammatory injury or of neoplastic cell fates has not been fully articulated. Regardless, the metaplastic cell fate is critical, as it is hijacked by mutant Kras in the development of pancreatic intraepithelial neoplasia (PanIN) and pancreatic ductal adenocarcinoma (PDAC).^8–12 Preclinical models have demonstrated sufficiency of mutant Kras and p53 to recapitulate the full spectrum of human disease from tumor initiation to metastasis.¹³ In the adult mouse, however, inflammation is essential, and the synergy between mutant Kras^G12D and inflammation to rewire chromatin drives neoplastic commitment.^8,9,11 Indeed, metaplasia does not fully resolve over time and can continues to support tumorigenesis despite temporal separation of inflammation and oncogenic Kras^G12D activation^14–15, highlighting the instrumental role of injury in driving lineage plasticity and oncogenic cell fate decisions.

Mutant KRAS, found in 95% of PDAC, is essential to both tumor initiation and maintenance^16–17, with KRAS^G12D the most common (35% of patients) and well-studied mutant allele. Similar to KRAS^G12D, KRAS^G12V, occurring in 30% of patients, requires in acinar-cell specific models an inflammatory insult to drive tumor formation⁸, but the degree to which epigenetic reprogramming is recapitulated in this mutational context has not been explored. Finally, early studies suggest that KRAS^G12R, which occurs in 15% of PDAC, is diminished relative to KRAS^G12D in its oncogenic potential.^18–19 Indeed, the precise KRAS mutation has been recently shown to powerfully influence clinical outcome and tumor features in human disease^20–21, such that the distinct biology of the Kras mutants is imperative to articulate, especially given emergent approaches to target mutant KRAS.^22–23

Herein we evaluate the molecular dynamics in injury and across the three most common mutant Kras alleles (G12D, G12V, and G12R) in tumor initiation, leveraging in each lineage-tracing models of adult Kras activation to interrogate specific cellular compartments, epigenetic reprogramming, and oncogenic signaling. Whereas Kras^G12D can robustly reprogram enhancers and drive AP-1 chromatin accessibility along with lineage reversion, Kras^G12R/V do not initiate the key cell-intrinsic epigenetic progenitor programs. Mechanistically, we tie this failure to defects in EGFR signaling, licensing of Pou2f3 and Vav1 in chromatin, and AKT and RAC1 activation, with only constitutive activation of PI3K/AKT signaling able to support tumor initiation in KRAS^G12R. Together, our data reveal the molecular basis of divergence among the Kras mutants, localize the phenotypic separation as occurring not with lineage plasticity but in neoplastic commitment, and establish the presence of mutation-specific oncogenic trajectories in pancreatic cancer.

RESULTS

Injury induces lineage reversion to a dedifferentiated cell state.

Acinar metaplasia in response to damage is traditionally explained as a transdifferentiation, as lineage-traced metaplastic lesions adopt a ductal morphology and expression of ductal-specific markers.^1–3 Metaplastic lesions have been demonstrated to contain cells that also express markers associated with pancreatic progenitor cells.^1,3–5,24 To benchmark the metaplastic response to early and late pancreatic progenitor cell fates, we generated Ptf1a-tdTomato (Ptf1a-tdT) mice to carry a knock-in allele containing both the tdTomato reporter and a functional copy of Pt1fa, which safeguards endogenous expression levels to ensure normal pancreatic development (Fig. S1A). Ptf1a-tdT mice thus display normal gross pancreatic morphology and tdTomato expression in adulthood (Fig. S1B). In addition, we confirmed exclusive labeling of the acinar compartment by immunofluorescent (IF) staining of tdTomato and GFP in adult Ptf1a^tdT/+; Pdx1^GFP/+ mice (Fig. S1C–D). Cells were sorted from dissociated Ptf1a-tdT embryos at e10.5 (early progenitors) and e15.5 (late progenitors) (Fig. S1E). In parallel, we sorted ductal cells from acinar cells derived from C57BL/6 mice via Cd49F and Cd133 staining (Fig. S1E), with RT-qPCR confirming appropriate discrimination of these cell types (Fig. S1F). Next, we performed RNA-seq on early progenitor, late progenitor, ductal cells, and acinar cells (the latter sorted from Mist1-Cre^ERT2; CAGs-LSL-tdTomato [MT] mice) finding that early progenitors were most distinct, while late progenitors represented an intermediate cell fate most like the differentiated cell types (Fig. S1G). These data were integrated to generate gene sets corresponding to each of the cell types and ‘exocrine’ and ‘progenitor’ cellular states (Mendeley Data Fig. 1).

Figure 1. Injury, not mutant Kras, can initiate enhancer reprogramming in the acinar compartment.

(A) PCA of ATAC-seq data for the top 5000 differentially accessible regions (DARs) in FACS-sorted early progenitor, late progenitor, acinar, ductal, and metaplastic (ADM) cells. (B) Heatmap of DARs identified after comparing ADM cells versus normal acinar cells, shown alongside late progenitor, ductal, and acinar cells. [log2(fold change)] > 0; padj < 0.01. Each column is a biological replicate mouse. (C) Barplot quantifying the number of DARs between ADM and various benchmark cell types (acinar, ductal, early and late progenitors). [log2(fold change)] > 0; padj < 0.01. (D) Venn diagram showing the overlap of DARs between ADM, late progenitor, and ductal cells, each compared to acinar cells. [log2(fold change)] > 1; padj < 0.01. (E) HOMER analysis depicting top motif enrichments when comparing ADM versus acinar. (F) Gene Ontology (GO) analysis using the Genomic Regions Enrichment of Annotations Tool (GREAT) on the regions with increased chromatin accessibility in ADM cells. (G) Schematic representation of the lineage-traced MK^G12DT mouse model. (H) Hematoxylin and eosin (H&E) staining of MK^G12DT mouse pancreas sections after 10 days (10d), 21 days (21d) and 7 weeks (7w) of mutant Kras activation. Representative images of n=3–5 mice/condition. Scale bar, 100μm. (I) Heatmap of DEGs identified after comparing sorted tdTomato+ WT and MK^G12DT cells from mice 10d and 7w after mutant Kras activation. [log2(fold change)] > 0; padj < 0.01. Each column is a biological replicate mouse. (J) GSEA of RNA-seq data comparing sorted tdTomato+ WT and MK^G12DT cells from mice 7w after mutant Kras activation. (K) Heatmap of DARs identified between sorted tdTomato+ WT and MK^G12DT cells from mice after 7w of mutant Kras activation. [log2(fold change)] > 0; padj < 0.01. Each column is a biological replicate mouse. See also Figures S1–S2.

To evaluate the molecular changes elicited with pancreatitis against these benchmarks, we turned to MT mice treated with caerulein (Fig. S1H). As expected, caerulein-treated mice showed signs of severe tissue damage with both stromal activation and advanced metaplastic lesions (Fig. S1H). Sorted ADM cells (tdT+ 2d after caerulein) were subjected to RNA-sequencing. Principal component analysis (PCA) demonstrated ADM largely consists of a ductal transition (Fig. S1I), with short Euclidean distances to both ductal and late progenitor types (Fig. S1J). Using the baseline molecular profiles generated from the normal cell types, we observed that ADM featured enrichment of our progenitor and specifically late progenitor gene sets (Fig. S1K). In this way, ‘ADM’ is not solely a ductal transition, but emerges from an unbiased developmental perspective as also an acinar cell dedifferentiation to a late progenitor state.

Enhancer reprogramming is induced by injury but not mutant Kras^G12D.

To understand epigenetic reprogramming elicited by injury, we sorted ADM, acinar, ductal, and early and late progenitor cells and performed the Assay for Transposase Accessible Chromatin-sequencing (ATAC-seq). PCA analysis revealed that, from a chromatin standpoint, ADM represents a hybrid ductal and late progenitor state (Fig. 1A). Interrogating all differentially accessible regions (DARs) between normal acinar and ADM cells, we observed strong commonalities of ADM with late progenitor and ductal samples (P1/P2 and D1 clusters, respectively), while also retaining some acinar features (A1/A2 clusters) (Fig. 1B). ADM cells had the fewest DARs compared to acinar cells (n=19,251) and late progenitors (n=46,975), indicative of their dedifferentiated chromatin state (Fig. 1C). In specific, ADM featured the robust acquisition of 5,590 peaks found in both late progenitor and ductal cell types (Fig. 1D) with strong enrichment of AP-1 motifs in gains of accessibility (Fig. 1E) and Gene Ontology (GO) terms corresponding to MAPK signaling (Fig. 1F).

Prior studies suggest that mutant Kras cooperates with injury to drive chromatin alteration in the pre-neoplastic pancreas.^5,12 These studies, however, have leveraged embryonic Kras activation wherein robust progression to neoplasia is observed with Kras alone and superimposed inflammation is dispensable.^24–25 In adult mice, mutant Kras alone is weakly tumorigenic, requiring prolonged activation to form sporadic neoplastic lesions and caerulein for robust tumorigenesis.^9,24,26–27 To examine whether mutant Kras can rewire chromatin in the adult mouse pancreas, we used MT; LSL-Kras^G12D (MK^G12DT) mice to activate mutant Kras^G12D in differentiated acinar cells (Fig. 1G). In keeping with prior reports^27–28, we observed only sporadic acquisition of ADM/PanIN up to 7 weeks after mutant Kras activation (Fig. 1H).

RNA-sequencing of sorted tdT⁺ cells at 10 days and 7 weeks revealed that Kras^G12D expression results in 109 and 551 differentially expressed genes, respectively (Fig. 1I). GSEA identified induction of Hallmark IFN-γ, IFN-α, and allograft rejection pathways, as well as EMT (Fig. 1J), with GO terms corresponding to immune and inflammatory responses (Fig. S2A). Despite these alterations to gene expression, there were only 18 DARs after 7 weeks of mutant Kras (Fig. 1K), and previously described Kras-induced DARs¹² were not dynamic in our model (Fig. S2B). These data suggest that Kras-induced transcriptional responses are not driven by direct changes in chromatin accessibility but by upregulation of pre-existing genes, activation of existing transcription factors, or post-translational modifications.

To determine if enhancers are dynamic in activation states rather than accessibility, we performed CUT&TAG on sorted acinar-derived cells for the histone modifications H3K4me1 and H3K27ac. Surprisingly, we found no clear clustering of WT and G12D samples in PCA analysis (Fig. S2C), and there were only 32 differentially modified regions for H3K27ac and 0 for H3K4me1. Querying the same set of Kras^G12D DARs reported previously¹², we observed that there were no gains in either K4me1 or K27ac (Fig. S2D). Both WT and Kras^G12D displayed strong correlations with one another (Fig. S2E), with no changes to either mark at key lineage-defining loci (Krt19, Ptf1a, Sox9, Pou2f3, Try4/5) (Fig. S2F). As mutant Kras alone is not sufficient to trigger robust epigenetic remodeling nor enhancer activation, these data highlight the dependency of mutant Kras on chromatin alterations that are induced by injury.

Kras^G12V and Kras^G12R are weak drivers of histologic transformation in the pancreas.

In pancreatic cancer, G12D mutation correlates with poorer survival, while the G12V and G12R mutations are associated with better outcomes.^{20–21,29–30} Embryonic Kras^G12R activation in vivo has been shown to be deficient as a potent inducer of premalignant disease¹⁸, but it is unclear how Kras^G12V and Kras^G12R compare to the well-studied Kras^G12D allele in tumor initiation. To systematically interrogate how the three most common Kras variants drive tumorigenesis, we generated MK^G12VgeoT and MK^G12RT mice (Fig. 2A). At 10 days, 21 days, and 7 weeks following mutant Kras activation, only Kras^G12D led to sporadic ADM and PanIN (Fig. 2B–C). At 4 months, Kras^G12D progressed to exhibit numerous and some advanced neoplastic lesions, which remained far fewer in Kras^G12V and Kras^G12R (Fig. 2D–F).

Figure 2. Kras^G12V and Kras^G12R are weak drivers of histologic transformation in the pancreas.

(A) Schematic representation of the lineage-traced MK^G12DT, MK^G12VgeoT, and MK^G12RT [MK^G12D/V/RT] mouse models after 10d, 21d, and 7w of mutant Kras activation. (B) H&E staining of MKT mouse pancreas sections collected after 10d, 21d, and 7w of mutant Kras activation. Representative images of 2–5 mice per condition. Scale bar, 100μm. (C) Histologic quantification of tissue area in pancreata collected from MK^G12D/V/RT mice after 10d, 21d, and 7w of mutant Kras activation. (D) Schematic representation of the lineage-traced MT (WT) and MKT mouse models after 4 months of mutant Kras activation. (E) H&E and Alcian blue/Sirius red staining of pancreas sections collected from WT and MK^G12D/V/RT mice after 4 months of mutant Kras activation. Images are representative of n=1–6 mice per condition. Scale bar, 200μm. (F) Histologic quantification of tissue area in pancreata collected from WT and MKT mice after 4 months of mutant Kras activation. Data are summarized as mean ± SEM. Two-way ANOVA was performed. (G) PCA of RNA-seq data for the top 2000 DEGs in tdTomato+ cells isolated from WT and tdTomato+ cells isolated from MKT mice after 7w of mutant Kras activation. (H) Bhattacharyya distance analysis of the top 2000 DEGs in tdTomato+ cells isolated from WT and MK^G12D/V/RT mice after 7w of mutant Kras activation. (I) Barplot quantifying the number of DEGs in MK^G12D/V/RT mice after 10d and 7w of mutant Kras activation. [log2(fold change)] > 1; padj < 0.01. (J) Heatmap of DEGs identified after comparing tdTomato+ cells isolated from WT and MK^G12D/V/RT mice after 7w of mutant Kras activation. [log2(fold change)] > 0; padj < 0.01. Each column is a biological replicate mouse. (K) GSEA performed on RNA-seq data from sorted tdTomato+ cells collected from MK^G12D/V/RT mice after 10d and 7w, using the indicated adult and progenitor pancreas benchmark gene sets.

Across these models, we isolated tdTomato-positive cells at 10 days and 7 weeks and performed RNA-seq. PCA showed minimal separation at 10 days (Mendeley Data Fig. 2), and clear separation of Kras^G12D mutant acinar-derived cells from the other alleles by 7 weeks (Fig. 2G–H). By 7 weeks, Kras^G12R was not able to maintain any DEGs as compared to Kras^WT (Fig. 2I–J), despite GSEA showing that Kras^G12R induced progenitor gene expression at 10 days most strongly (Fig. 2K). By the 7 week timepoint, only Kras^G12D robustly induced acquisition of ductal and late progenitor transcriptional modules and loss of acinar identity programs (Fig. 2K). These findings suggest that Kras^G12R can initiate but not sustain mutant Kras signaling, with Kras^G12V as an intermediate potency mutant.

Lineage plasticity is preserved in Kras^G12V and Kras^G12R mutants following injury.

Given the requirement for inflammation to induce robust tumor initiation, we next asked how injury facilitates lineage plasticity across the Kras mutants. We induced mutant Kras and pancreatitis using tamoxifen and caerulein, respectively and interrogated mice at the peak of injury (2 days) and 21 days later (Fig. 3A). At the peak of injury (2d), all caerulein-treated Kras mutant mice exhibited robust formation of ADM (Fig. 3B) and induction of SOX9 and CK19 expression in metaplastic lesions (Fig. S3A), to a degree greater than Kras^WT and indistinguishable between the three Kras mutants (Fig. 3C).

Figure 3. Lineage plasticity is not diminished in Kras^G12V and Kras^G12R mutants.

(A) Schematic representation of the lineage-traced MK^G12D/V/RT mouse models with acute injury following by 2 days (2d) and 21 days (21d) of recovery. (B) H&E staining of WT and MK^G12D/V/RT mouse pancreas sections collected 2d post-acute injury. Representative images of n=4–5 mice/condition. Scale bar, 100μm. (C) Histologic quantification of tissue area in pancreata collected from WT and MK^G12D/V/RT mice 2d post-acute injury. ADM lesions further subdivided based on severity (early and advanced). (D) PCA of RNA-seq data for the top 2000 DEGs in tdTomato+ cells isolated from WT, ADM (WT + injury) and MK^G12D/V/RT mice collected 2d post-acute injury. (E) Bhattacharyya distance analysis of the top 2000 DEGs in tdTomato+ cells isolated from WT, ADM (WT + injury) and MK^G12D/V/RT mice 2d post-acute injury. (F) Heatmap of all DEGs identified after comparing tdTomato+ cells isolated from WT, ADM (WT + injury) and MK^G12D/V/RT mice 2d post-injury. [log2(fold change)] > 1; padj < 0.01. Each column is a biological replicate mouse. (G-H) GSEA performed on RNA-seq data from tdTomato+ cells from MK^G12D/V/RT-derived tdTomato+ cells 2d post-acute injury, using adult and developing pancreas benchmark gene sets, with comparison to WT + injury (G) and between KRAS alleles (H). (I) Heatmap of all DARs identified after comparing WT with MK^G12DT cells 2d post-acute injury. [log2(fold change)] > 0; padj < 0.01. Each column is a biological replicate mouse. See also Figure S3.

To define how peak injury and oncogenic Kras synergize to drive lineage plasticity, we isolated acinar-derived cells in WT, Kras^G12D, Kras^G12V, and Kras^G12R contexts, 2 days after caerulein. PCA of RNA-seq showed minimal separation of Kras^G12D (Fig. 3D), and Bhattacharyya distances between the three mutants were quite small (Fig. 3E). Similarly, Kras mutants were highly similar across the ~6000 DEGs, with the most dramatic differences between all injury contexts (WT and G12X) and control acinar cells (Fig. 3F). While all three mutants were able to upregulate ductal transcripts, G12D was able to do so to a greater degree, and could also drive re-expression of the early progenitor module and to repress acinar cell identity, with G12V and G12R retaining the late progenitor features associated with ADM (Fig. 3G–H). Gene pattern analysis showed that most DEGs were either induced or repressed in the context of ADM and the degree of change was mildly enhanced in the context of Kras^G12R, more markedly altered in Kras^G12V, and most significantly changed in Kras^G12D (Fig. S3B), defining a hierarchy amongst the three alleles.

To understand how Kras^G12D and Kras^G12R diverge molecularly at the peak of injury despite similar histology, we carefully interrogated differences in gene expression between these alleles. Along with several MAPK pathway genes (Mapk3 [ERK1], Mapk11, Mapk6, Map2k1 [MEK1], we observed that Klf5, implicated in the progenitor program induced by injury⁵, and IL-33, a nuclear cytokine that supports tumorigenesis¹², were induced in G12D but not in G12R (Fig. S3C). Conversely, Bhlha15 (Mist1), a key acinar identity TF, was retained in G12R (Fig. S5C). Pathway analysis showed enrichment of KRAS signaling, TGF-β, and EMT gene sets in G12D (Fig. S3D). Finally, enriched GO terms in G12D corresponded to actin filament / cytoskeletal organization processes (Fig. S3E), aligning with enrichment of the apical junction Hallmark in G12D (Fig. S3D), suggesting a diminished capacity for G12R to alter epithelial polarity and the actin cytoskeleton in tumor formation.

Next, we examined how the different Kras alleles are able to rewire chromatin in the context of caerulein. Kras^G12D most markedly drove gains of accessibility amongst gains shared with Kras^G12R and Kras^G12V (Fig. 3I; cluster 2) but also featured an allele-specific cluster of gains in ATAC signal (Fig. 3I; cluster 3). As expected, ATF3 and AP-1 motifs were enriched in Kras^G12D-specific gains of accessibility (Fig. S3F). These data thus highlight the unique ability of Kras^G12D to drive tumorigenic epigenetic and transcriptional programs. Indeed, at 21 days after injury, the histology of Kras^G12D showed widespread PanIN and Kras^G12V giving rise to occasional PanIN, with Kras^G12R largely indistinguishable from WT (Fig. S3G–H), reflecting the ultimate phenotypic consequence of the molecular differences observed at 2 days.

Kras^G12D is unique in its ability to drive robust neoplastic transformation.

As initial lineage plasticity was comparable between Kras mutants immediately following acute caerulein, we asked if prolonged injury could sustain neoplastic transformation in Kras^G12V and Kras^G12R. To this end, we administered caerulein using a 3-week protocol and harvested the pancreas at either peak injury (2 days), 21 days, or 3 months (Fig. 4A). Surprisingly, prolonged inflammation could not support Kras^G12V or Kras^G12R tumorigenesis (Fig. 4B), with a striking separation of G12D that was enhanced over time (Fig. 4C). In survival analyses, G12D mice uniformly did not survive beyond 24 weeks (owing to pancreatic insufficiency), whereas WT, G12V, and G12R mice generally survived >1 year (Fig. 4D).

Figure 4. Kras^G12D is unique in its ability to drive robust neoplastic transformation.

(A) Schematic representation of the lineage-traced MT and MK^G12D/V/RT mouse models after prolonged injury, as well as 2 days (2d), 3 weeks (3w), and 3 months (3m) of recovery. (B) H&E staining of MT and MK^G12D/V/RT mouse pancreas sections collected 2d, 3w, and 3m post-prolonged injury. Alcian blue/Sirius red staining is also shown at 3m. Representative images of n=3–5 mice/condition. Scale bar, 100 μm. (C) Histologic quantification of tissue area in pancreata collected from MT and MK^G12D/V/RT mice 2d, 3w, and 3m post-prolonged injury. (D) Kaplan-Meier curves of overall survival for MT and MK^G12D/V/RT mice after prolonged injury. (E) PCA of RNA-seq data for the top 2000 DEGs in tdTomato+ cells isolated from WT mice (w/o injury) and WT and MK^G12D/V/RT mice collected 3w post-prolonged injury. (F) Heatmap of all DEGs in sorted cells form WT mice (no injury) and MK^G12DT mice 3w post-prolonged injury. [log2(fold change)] > 1; padj < 0.01. Each column is a biological replicate mouse. (G) PCA of ATAC-seq data for the top 2000 DARs in tdTomato+ cells in sorted cells form WT mice (no injury) and WT/MK^G12D/V/RT mice 3w post-prolonged injury. (H) Venn diagram showing the overlap of DARs identified by comparing tdTomato+ cells isolated from both MK^G12DT and MK^G12VgeoT collected 3w post-prolonged injury. [absolute log2(fold change)] > 1; padj < 0.01. (I) GSEA performed on RNA-seq data from tdTomato+ cells isolated from WT mice (no injury) and MK^G12D/V/RT 3w post-prolonged injury, using adult and developing pancreas benchmark gene sets. See also Figure S4.

As before, we sorted acinar-derived cells from WT, Kras^G12D, Kras^G12V, and Kras^G12R mice 3 weeks after the completion of prolonged injury. In this context, G12D transcriptionally was quite different from the other alleles, both in PCA (Fig. 4E) and when interrogating DEGs in the set (Fig. 4F). When evaluating the ATAC-seq data, PCA again showed that the G12D mutant separates away from the other Kras mutant samples and control samples (Fig. 4G), with 39,632 DARs as compared to 2,912 for G12V and 0 for G12R (Fig. 4H). At the pathway level, all three mutant Kras alleles could drive expression of ductal and progenitor gene modules (Fig. 4I), suggesting that lineage reversion and transdifferentiation programs were uniformly inducible. Similarly, Kras^G12R and Kras^G12V were able to drive EMT and inflammatory signaling pathways (Fig. S4A). Notably, however, Kras^G12R and Kras^G12V were deficient in their ability to repress exocrine / acinar cell fate (Fig. 4I). Kras^G12D was unique in its ability to unveil TNF signaling via NF-κB and apical junction Hallmarks (Fig. S4B). GO term analysis on DEGs supported the distinct ability of G12D to drive cytoskeletal / actin filament organization (Fig. S4C). In keeping with these findings, we observed in the ATAC data that G12D and to a lesser extent G12V, but not G12R, were able to induce PDAC-associated epigenetic reprogramming (Fig. S4D). Together, these data define a phenotypic defect in G12R, and to a lesser extent in G12V, not to change identity but to sustain specific oncogenic transcriptional and epigenetic programs despite a prolonged inflammatory insult.

Enhancer reprogramming in Kras^G12D features activation of the neoplastic determinants Pou2f3 and Vav1.

To understand enhancer reprogramming elicited by the pairing of Kras^G12D with caerulein-mediated injury, we performed CUT&TAG for H3K4me1 and H3K27ac on sorted acinar-derived cells. Distinct from our findings absent injury, we observed that G12D + injury dramatically shifts both enhancer priming (H3K4me1) and activation (H3K27ac) states (Fig. 5A). Injury alone showed a sustained impact over time on H3K4me1 but not H3K27ac, in keeping with our previous report on the epigenetic encoding of inflammatory memory.¹⁴ Comparing G12D with G12R, we observed thousands of chromatin domains that were differentially modified (Fig. 5B). G12R, in particular, displayed persistent accessibility and K4me1 and K27ac at digestive enzyme loci (Cel, Cela2a, Try5, Cpa1, etc.) (Fig. 5C). Unsurprisingly, HOMER analysis showed G12R enriched for bHLH motifs (i.e. Ptf1a, Mist1) (Fig. 5E), while chromatin with higher accessibility / K4me1 / K27ac in G12D was enriched for bZIP (i.e. AP-1), ETS, and POU family motifs (Fig. 5E). A multimodal analysis of the most dynamic chromatin and gene expression in G12D as compared to G12R revealed the tuft cell lineage-defining TF Pou2f3 as an outlier, displaying significantly increased gene expression, accessibility, H3K4me1, and H3K27ac in G12D versus G12R (Fig. 5F). Indeed, these differences were highly consistent across biological replicates in enhancers within Pou2f3 as well as Vav1 (Fig. S5A–C), a tuft cell marker and key guanine nucleotide exchange factor (GEF) for the Rho GTPase RAC1³², which in turn activates PAK1 and p38/JNK.

Figure 5. Enhancer reprogramming in Kras^G12D features activation of the neoplastic determinants Pou2f3 and Vav1.

(A) PCA of CUT&TAG data for H3K4me1 and H3K27ac in tdTomato+ cells isolated from WT without injury and tdTomato+ cells isolated from MT and MK^G12D/V/RT mice collected 3w post-prolonged injury. (B) Tornado plots of differentially modified regions (DMRs) for H3K4me1 and H3K27ac as determined by CUT&TAG for MK^G12DT and MK^G12RT mice. Each tornado represents a unique biological replicate. (C) Volcano plots of differentially accessible (left panel), H3K4me1-enriched (middle panel), and H3K27ac-enriched (right panel) regions between MK^G12DT and MK^G12RT sorted tdTomato+ cells. (D-E) HOMER analysis depicting enriched motifs for G12R (D) and G12D (E) for the indicated features (ATAC, K4me1, K27ac). (F) Scatter plot of all genomic regions in G12D versus G12R for ATAC signal (x-axis), K4me1 (y-axis), K27ac (color), RNA levels (point size). (G-H) Immunohistochemical (IHC) staining of pancreatic sections for POU2F3 in (G) MK^G12D/V/RT mice 3w after prolonged injury, together with IF for DCLK1, and (H) MK^G12RT mice 3m and 12 months (12m) after prolonged injury. Representative images of n=2–5 mice/condition. Scale bar, 100μm. See also Figure S5.

Kras^G12R and Kras^G12V feature impaired EGFR signaling.

Given a clear hierarchy amongst the PDAC-associated Kras variants, we asked why Kras^G12R and Kras^G12D drive markedly different molecular and phenotypic outcomes. Using the Cytokine Signaling Analyzer (CytoSig)³¹, we observed that EGF and TGF-α family signaling are strongly induced in G12D but not as well in the other allelic contexts (Fig. S6A). Examination of the most enriched cytokine transcripts nominated amphiregulin (AREG) and epiregulin (EREG) as key EGFR ligands that are G12D-specific in acute injury and more highly enriched as phenotypes diverge with prolonged injury (Fig. S6B). To confirm that these ligands are elevated in G12D at the protein level, we performed a multiplexed cytokine array that revealed a 50-fold induction of AREG in G12D, compared to ~2-fold in G12R/V (Fig. S6C). Importantly, several key cytokines elaborated in PDAC initiation, including IL-6, LIF, and Cxcl1/2, were also higher in G12D (Fig. S6D).

To ascertain if the difference in Kras mutants could be ‘rescued’ by EGFR ligand administration, we tested the capacity of the mutants to undergo ADM in vitro^1,13 using recombinant human TGFα administered to acinar explants. However, in spite of excess of EGFR ligand, we observed more robust ductal cyst formation in Kras^G12D acinar cells than in Kras^G12V/R (Fig. S6E–F). Mutant KRAS^G12D is known to drive increased expression of EGFR itself³³, and, indeed, neoplastic transformation is known to depend on the presence of signal through the receptor.^33–34 To determine if EGFR upregulation can be induced by G12V/R, we performed Western blotting, and observed that both EGFR and its phosphorylation at Y1068 (but not Y992) were inducible in G12D but not in G12V/R (Fig. S6G–H). Given defects both in EGF ligand and receptor accumulation in G12V/R, we administered the EGFR inhibitor afatinib across the mutants, observing that absent EGFR signaling, G12D resembles G12R (Fig. S6I). Collectively, these data support a multi-level defect in G12V and G12R to drive both EGF ligand and receptor activation that are necessary for KRAS-mediated oncogenic transformation.

Activation of Rac1 signaling is deficient in the Kras^G12R mutant pancreas.

Next, we evaluated the signaling implications of dampened EGF receptor-ligand interactions at the plasma membrane. Because Kras^G12R has been described to have different GTP hydrolysis activity^35–36, we examined RAS-GTP levels in the mouse pancreas 7 weeks after tamoxifen injection. All three alleles increased active RAS to comparable degrees above wild-type, with minimal difference (Fig. 6A; Fig. S7A). ERK phosphorylation, notably, was elevated in vivo in G12D as compared to G12R, with G12V representing an intermediate phenotype (Fig. S7B); however, by IHC, G12R showed a capacity to generate p-ERK⁺ PanIN, albeit with significant temporal delay (Fig. 6B), suggesting that MAPK signaling is delayed but not deficient in G12R. On the other hand, AKT phosphorylation was largely absent in both G12R and G12V by IHC (Fig. 6C), suggesting an inability altogether to activate this arm of oncogenic signaling.

Figure 6. Activation of Rac1 signaling is deficient in the Kras^G12R mutant pancreas.

(A) Ras-GTP levels from bulk pancreas tissue lysate extracted from MT and MKT mice 7w post-tamoxifen injection. (B) IHC staining of pancreatic sections for phospho-Erk in both MK^G12DT and MK^G12RT mice 3w, 3m, and 12m after prolonged injury. Representative images of n=2–5 mice/condition. Scale bar, 200μm. (C) IHC showing Akt (S473) phosphorylation on pancreas sections from MT, MK^G12D/V/RT mice 3w post-prolonged injury. Scale bar, 100μm. (D) Reverse-phase protein array (RPPA) analysis of differentially abundant proteins between MK^G12DT and MK^G12RT 3w after prolonged injury; n=3 mice/condition. (E) GSEA performed on RNA-seq data from tdTomato+ cells isolated from WT without injury and MK^G12D/V/RT mice 3w post-prolonged injury. (F) Rac1-GTP levels from bulk pancreas lysate extracted from MT and MK^G12D/V/RT mice 3w after prolonged injury. (G) Western blot for VAV1 on pancreas lysates obtained from sorted acinar-derived (tdTomato+) cells from MT and MK^G12D/V/RT mice 3w after prolonged injury. (H) IHC staining of pancreatic sections for VAV1 in both MK^G12DT and MK^G12RT mice 3w, 3m, and 12m after prolonged injury. Representative images of n=2–5 mice/condition. Scale bar, 200μm. (I) Western blot for TIAM1 and VAV1 on sorted tdTomato+ cells collected from MT and MK^G12D/V/RT mice 7w post-tamoxifen injection. See also Figures S6–S7.

To ascertain in an unbiased fashion the signaling differences between Kras^G12D and Kras^G12R, we performed a reverse-phase protein array (RPPA) on bulk pancreas lysates obtained 3 weeks following prolonged injury (as in Fig. 4A). We observed that VAV1 and PAK1 were enriched in G12D (Fig. 6D), at ~40-fold and ~6-fold, respectively, more than G12R. Our GSEA and GO analyses suggested a G12R loss of apical junction and actin remodeling (Fig. S4B–C), processes in which VAV1-dependent RAC1 signaling is essential and that are required for PDAC initiation.^37–38 Indeed, we observed in the RNA-seq data that gene sets related to Rac1 signaling pathways, but not gene sets relating to MAPK/ERK signaling nor the recently developed KRAS molecular signature³⁹, were enriched in G12D relative to G12R and G12V (Fig. 6E). Given the well-established role for RAC1 activity in key cytoskeletal processes and PAK activation, our data pointed overall to the AKT and RAC1 pathways as the key signaling modules different between G12D and G12R.

Next, we directly queried if active RAC1 was different between Kras alleles. In bulk pancreas lysates, using a GST-PAK1-protein binding domain(PBD), we detected much higher RAC1-GTP in G12D than in G12V or G12R (Fig. 6F; Fig. S7C). We then validated the observed difference between G12D and G12R in VAV1 abundance (found in the RPPA) by Western blotting (Fig. 6G; Fig. S7D). As Vav1 is expressed highly in the hematopoietic compartment, we confirmed by IHC that the G12D pancreatic epithelium was VAV1⁺, as in prior reports⁴⁰, and that this staining was confined largely to PanIN lesions. In addition, the G12R pancreas was largely devoid of VAV1 staining, and PanIN that eventually emerged failed to stain robustly for VAV1 (Fig. 6H). Finally, we asked if VAV1 differences across mutant Kras alleles preceded phenotypic divergence between the alleles. Indeed, VAV1 abundance showed marked difference between G12D and G12R in the context of Kras activation alone (at 7 weeks) (Fig. 6I; Fig. S7E) as well as in sorted tdTomato⁺ cells, indicating that the VAV1 difference is established early in the acinar compartment (Fig. S7F). However, in the context of acute injury, wherein the Kras alleles produce similar phenotypes (Fig. 3B), we observed no difference in VAV1 abundance (Fig. S7G), suggesting that VAV1 can initially be articulated in G12R in initial lineage plasticity, but cannot be maintained over time (Fig. 6H). Importantly, VAV1 has been shown to be dependent on EGFR⁴⁰, connecting the deficient epigenetic reprogramming in G12R that fails to activate Vav1 in chromatin to the upstream inability to initiate the circuit of autocrine AREG upregulation and EGFR activation and to the downstream diminished RAC1 signaling. Our data thus converge on the notion that VAV1 - previously described to be a key node in pancreatic tumorigenesis^40–42, and also lost in the absence of p11α⁴¹ – is not activated in the context of Kras^G12R, and that the failure of G12R to mediate oncogenic transformation is due to a lack of EGFR/AKT/VAV1 supporting sustained RAC1 activity.

Constitutive activation of AKT can support Kras^G12R-driven tumorigenesis.

In light of the inability of Kras^G12R to drive robust tumorigenesis, we aimed to determine if this inability could be overcome by secondary perturbations in vivo. We turned to the well-established ‘KPC’ model of PDAC featuring p48-Cre and mutant p53 (LSL-Trp53^R172H) to evaluate if tumor suppressor loss could ‘rescue’ Kras^G12R (Fig. 7A). Surprisingly, at 20 weeks of age, the G12R mouse pancreas was devoid of lesions (Fig. 7B) at a time point when G12D mice already developed robust PanIN. Mouse survival was substantially prolonged (Fig. 7C), but beginning at approximately 40 weeks of age, several G12R/p53^mut mice developed lymphomas or required euthanasia for other conditions. In total, of the 11 G12R mice that died in the course of the study, 6 had evaluable pancreas histology, none of which contained PanIN lesions or PDAC. In contrast, all G12D mice did not survive >40 weeks. We suspected that biallelic inactivation of p53 might potentiate neoplastic commitment, and generated MK^G12RT; LSL-Trp53^R172H/flox mouse models accordingly (Fig. 7D). Strikingly, whereas G12D;p53^mut/flox animals rapidly developed PanIN and PDAC by 20 weeks of age (Fig. 7E), G12R;p53^mut/flox animals were free of neoplastic lesions and showed robust overall survival (Fig. 7F).

Figure 7. Constitutive activation of AKT can support Kras^G12R-driven tumorigenesis.

(A) Schematic representation of p48-Cre; LSL-Kras^G12D/R; LSL-Trp53^R172H/+ mouse model. (B) H&E sections of mouse pancreas from animals as described in (A). Scale bar, 100μm. (C) Kaplan-Meier survival of mice as per (A); Mantel-Cox test was performed. (D) Schematic representation of the lineage-traced MK^G12D/V/RT mouse models with Trp53^R172H/flox alleles. (E) H&E staining of pancreas from animals as described in (D), at the indicated timepoints. Scale bar, 100μm. (F) Kaplan-Meier survival of mice as per (D); Mantel-Cox test was performed. (G) Schematic representation of the lineage-traced MK^G12RT +/− Rac1^P29S OR myr-Akt mouse models after acute injury. (H-I) H&E staining of (H) MK^G12RT +/− Rac1^P29S OR myr-Akt and (I) MT;myr-Akt +/− Kras^G12R mouse pancreas sections collected 3w post-acute injury. Scale bar, 100μm. (J) IHC staining for VAV1 and phospho-AKT in MT;myr-Akt +/− Kras^G12R mouse pancreas sections collected 3w post-acute injury. Scale bar, 100μm. (K) Rac1-GTP levels from bulk pancreas lysate from MK^G12D/V/RT mice 3w post-acute injury. (L) Scatter plot of the mutational frequency (%) of the indicated genes in KRAS^G12D (x-axis) and KRAS^G12R / KRAS^G12V tumors (y-axis), using the projectGENIE dataset. (M) Forest plot of the relative odds ratio (G12R/G12D) for co-occurring mutation, using the projectGENIE dataset.

Given the heterogeneity among KRAS mutants to interact with RAF kinases, we next asked if modulation of MAPK activation could ‘rescue’ the G12R defect. To test this, we perturbed MAPK activation in MK^G12RT mice through either administration of low-dose MEK inhibitor (trametinib; 0.5mg/kg) or RAF inhibitor (PLX4720; 472 mg/kg) to either diminish or paradoxically amplify^43–45 ERK signaling, respectively, addressing the possibility that Kras^G12R might lead to too much or too little signaling through this output (Fig. S7H). Tissue analysis after 3 weeks of the addition of trametinib or PLX-4720 to the diet, however, showed no significant PanIN development in G12R mice (Fig. S7I).

We turned our attention to the key signaling defects in G12R. As PI3K/AKT can also directly activate RAC1⁴⁰, we suspected that either constitutive activation of AKT or RAC1 could ‘rescue’ the ability to drive PanIN in the Kras^G12R context. To test this directly, we crossed a myristoylated AKT (myrAKT) allele⁴⁶ or the constitutively active Rac1^P29S allele⁴⁷ with MK^G12RT mice (Fig. 7G). While constitutive RAC1 activation failed to support tumor initiation, we found that enforcement of membrane localization of AKT could indeed rescue the G12R phenotype (Fig. 7H). As activating PI3K mutations have been shown to independently drive pancreatic tumorigenesis, even in the absence of mutant Kras⁴⁸, we confirmed that while myrAKT alone gave rise to ADM and scattered PanIN, the addition of KRAS^G12R dramatically potentiated neoplastic commitment (Fig. 7I). To determine if myrAkt expression can rescue the molecular defects of G12R in pancreatic tumorigenesis, we performed IHC on myrAkt and injury +/− Kras^G12R pancreases, finding that both VAV1 and p-AKT are restored in this allelic context (Fig. 7J). Active RAC1 is also restored in Kras^G12R + myrAkt mice (Fig. 7K), such that these data show that constitutive activation of AKT can cooperate with KRAS^G12R to rescue the molecular impairments in this allele to drive tumorigenesis.

In light of these findings, we suspected that human KRAS^G12R would display evidence of deficient oncogenic signaling and a requirement for PI3K/AKT/RAC1 activation. We queried the projectGENIE cohort of >8,000 PDAC patients to determine if G12V/G12R mutation co-occurs with other mutational or copy number alterations. While most mutational events, including to TP53, SMAD4, and CDKN2A, did not display a predilection for a specific KRAS allele (Fig. 7L), PIK3CA mutations were enriched in KRAS^G12R (Fig. 7L–M). Indeed, we observed a trend towards statistically significant enrichment of all PI3K pathway gene mutations, including PIK3CB and AKT2 (Fig. 7M). These data support the notion that in human disease there are defects in KRAS^G12R signaling that can be compensated for by activating mutations to the PI3K pathway; these findings align with the G12R mouse models that show an inability to support RAC1 activity, its key GEF VAV1, and AKT activation, all of which are downstream of the defective KRAS-PI3K interaction.

DISCUSSION

Cancer initiation is a complex process wherein mutational events in association with exogeneous insults can synergistically overcome biological barriers that maintain tissue homeostasis. In this study, we have employed models wherein KRAS mutations can be induced in the adult acinar cell, avoiding limitations of extant systems employing alternative drivers (such as Ptf1a-Cre^ER) that impact acinar cell differentiation⁴⁹ and lineage tracers (such as mTmG and rtTA-IRES-mKate [RIK]) that give rise to occult inflammation¹⁸. Employing this system, coupled with development of the Ptf1a-tdTomato, mouse, we have been able to broadly interrogate acinar-to-progenitor dedifferentiation, beyond the accumulation of specific individual markers.^1,5,50–53 Finally, the literature has largely employed Kras^G12D mutations as the canonical driver of PDAC; our work evaluates in parallel the oncogenic potential of Kras^G12V and Kras^G12R - representing nearly half of all PDAC cases – which are prognostic of patient outcome and biologically distinct in the human setting.²⁰ Through a combination of lineage-tracing models and molecular analyses, and reconstruction of the developmental trajectory of exocrine cells of the pancreas, our work has revealed that injury-induced ADM reflects a mixed-lineage state that features broad epigenetic reprogramming. Mutant Kras, unlike injury, induces oncogenic signaling and alterations to gene expression, but does not lead to any substantial rewiring of chromatin accessibility or enhancer states. Kras^G12D successfully co-opts the lineage reversion brought about by injury but the other mutants do not. Kras^G12R, in particular, can induce an initial degree of gene expression alteration that can support lineage plasticity, but does not sustain either a histologic transformation, transcriptional rewiring, nor epigenetic reprogramming of the tuft cell lineage transcription factor Pou2f3 and the tuft cell marker and RAC1 GEF Vav1.

There are a multitude of mechanisms by which Kras^G12D, Kras^G12V, and Kras^G12R mutants could give rise to varied phenotypes. Biochemical analyses have shown that G12D and G12V comparably lower the rate of GTP hydrolysis, and that G12R displays the most dramatic reduction as compared to WT KRAS³⁴, suggesting that KRAS^G12R might be more locked into its active conformation. However, the KRAS mutants also vary in their GEF interactions, with SOS1 catalytic constants (k_cat) lowest for G12R and GRP1 k_cat lowest for G12D.¹⁹ As SOS1 is the dominant GEF for KRAS, these data imply that G12R might instead be less ‘active’ than G12D by virtue of an inability to engage in rapid GTP-GDP exchange. In our data, we observe that RAS-GTP levels appear largely comparable between the Kras mutants, suggesting that altered GTPase and GTP/GDP exchange do not explain different phenotypic outcomes found herein.

KRAS mutants also differ in their interaction with downstream effectors. RAF affinity has been described as decreased in all three KRAS variants as compared to WT³⁵, and KRAS^G12R specifically causes a switch II domain structural change that affects its interaction with p110α.¹⁹ Importantly, Kras^G12D p110α-knockout mice produce a similar phenotype as Kras^G12R - an inability to sustain injury-induced metaplasia, a defect in VAV1 upregulation, and a lack of sustained RAC1 signaling.⁴¹ Our data thus suggest that the differences between the Kras mutants can thus be explained by differences in p110α-dependent RAC1 / VAV1 activation, pathways that are integral to cell migration and cytoskeletal dynamics, are dependent on EGF receptor-ligand interactions, and that are enriched in G12D but not the other alleles. However, constitutive activation of RAC1 signaling via the P29S mutant does not ‘rescue’ the G12R phenotype, but enforced AKT signaling does. Thus the G12R defect is multifactorial, consisting of a failure to induce VAV1, a loss of RAC1 activation, and inadequate proliferative signaling through AKT; importantly, we show that myrAkt overexpression can fix all of these specific impairments.

A key implication of our data extends to therapeutic targeting of KRAS in each allelic context, as KRAS mutants support the pleiotropic downstream effectors to varying degrees. Given that MAPK, PI3K, and RAC1 pathway activation are essential to tumor initiation, one would suspect that G12D, capable of supporting all three signaling outputs, might be more dependent on mutant KRAS than G12R. Indeed, the recent phase III clinical trial of the pan-RAS(ON) inhibitor daraxonrasib versus chemotherapy showed a significant reduction in the hazard ratio for death in patients with G12D/V (HR 0.38 [95% CI, 0.28–0.53]), compared to 0.70 (0.34–1.44) for other G12 mutations.⁵⁴ Whether this codon 12 mutant-specific variation in the response to KRAS targeting is driven by the distinct biology of G12R remains an open question.

Taken together, our work herein underscores the complex cell fate decisions that typify the responses to inflammatory injury and oncogenic stress. Whereas these two essential components conspire to drive oncogenesis - even with broad temporal separation¹⁴ – they can be viewed as serving unique, non-overlapping roles in this process. Injury induces the requisite epigenetic reprogramming that Kras cannot; Kras, in turn, can co-opt the mixed-lineage state to sustain a stable neoplastic cell fate decision featuring articulation of key neoplastic determinants such as Pou2f3 and Vav1. Moreover, whereas lineage plasticity is facilitated by all three Kras mutants, neoplastic commitment depends significantly on the precise mutation in Kras, suggesting that tumorigenesis in each mutational context has distinct allele-specific molecular dependencies, and perhaps therapeutic vulnerabilities.¹⁹ Quite clearly, KRAS allelic status can shape clinical outcomes in human pancreatic cancer by driving distinct biological behavior; here we demonstrate that the basis for differences in cancer begin with distinct evolutionary trajectories preceding tumorigenesis. Moving forward, these differences highlight the diverse biological implications of Kras mutations – across preclinical models, human disease, and pharmacologic targeting of KRAS -- that warrant further study to enable the rational and precise interception of pancreatic cancer.

Limitations of the study.

While our study employs several mouse models and overall large sample sizes to detect substantial differences across different KRAS mutations in the pancreas, there are nonetheless a number of limitations. First, this work was performed entirely in mouse models and the extent to which the divergent biology of KRAS alleles in these models recapitulates human disease is not entirely known. Moreover, while we find that enforced AKT activation can support Kras^G12R-driven tumorigenesis, how this barrier might be traversed in patients who do not harbor PIK3CA pathway mutations remains a key area of future inquiry. Third, the primary focus of this work has been the epithelial compartment; the degree to which stromal and immune compartments differentially restrain or permit lineage plasticity and neoplastic commitment in different KRAS mutants has not yet been articulated herein. Finally, our work identifies substantial barriers for initiation of Kras^G12R-mutant tumors in the pancreas, yet this organ represents the primary epithelial tissue in which Kras^G12R-mutant tumors are found – a paradox that is imperative to resolve.

RESOURCE AVAILABILITY

Lead contact.

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Rohit Chandwani (roc9045@med.cornell.edu).

Materials availability

Mouse lines (Ptf1a-tdTomato) generated in this study have been deposited to The Jackson Laboratory.

Data and code availability

All high-throughput sequencing data have been deposited at GEO and are publicly available. Additional data have been deposited at Mendeley and are publicly available. Accession numbers are listed in the key resources table. Microscopy data reported in this paper will be shared by the lead contact upon request.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit IgG Antibody, CUTANA™ CUT&RUN Negative Control	EpiCypher	13-0042; RRID:AB_2923178
Anti-Rabbit Secondary Antibody for CUTANA™ ChIC/CUT&Tag Workflows	EpiCypher	13-0047
Anti-Mouse Secondary Antibody for CUTANA™ ChIC/CUT&Tag Workflows	EpiCypher	13-0048; RRID:AB_3676529
Histone H3K27ac Antibody, SNAP-ChIP Certified	EpiCypher	13-0045; RRID:AB_2923489
Histone H3K4me1 Antibody, SNAP-ChIP Certified	EpiCypher	13-0040; RRID:AB_2923143
H3K27me3 Monoclonal Antibody (G.299.10), ChIP-Verified	ThermoFisher Scientific	MA5-11198; RRID:AB_11000749
Acetyl-Histone H3 (Lys27)	ThermoFisher Scientific	MA5-23516; RRID:AB_2608307
CD45 Recombinant monoclonal antibody (6019)	Proteintech	80297-1-RR; RRID:AB_2918884
P44/42 MAPK (Erk1/2) (137F5) Rabbit mAb	Cell Signaling Technology	4695S; RRID:AB_390779
Anti-Vinculin antibody [EPR8185] (ab129002)	Abcam	ab129002; RRID:AB_11144129
Goat Polyclonal anti-Chicken IgY H&L (Alexa Fluor® 488)	Abcam	ab150169; RRID:AB_2636803
Goat Polyclonal anti-Mouse IgY H&L (Alexa Fluor® 488)	Abcam	ab150113; RRID:AB_2576208
Goat Polyclonal Anti-Mouse IgG H&L (Alexa Fluor 647) Antibody	Abcam	ab150115; RRID:AB_2687948
Rabbit anti-human POU2F3 antibody	Millipore Sigma	av32537; RRID:AB_1855584
VAV1 Antibody #2502	Cell Signaling Technology	2502S; RRID:AB_2213556
Anti-VAV1 (phospho Y174) antibody [EP510Y] (ab76225)	Abcam	ab76225; RRID:AB_1524546
Mouse Amphiregulin antibody	R&D systems	AF989; RRID:AB_2060663
Phospho-EGFR (Tyr992) Polyclonal Antibody	Phospho-EGFR(Tyr992)	44-786G; RRID:AB_2533753
Phospho-EGF Receptor (Tyr1068) (D7A5) XP® Rabbit mAb	Cell Signaling Technologies	3777S; RRID:AB_2096270
EGFR Antibody (A-10)	Santa Cruz Biotechnology	sc-373746; RRID:AB_10920395
Phospho Akt (Ser473) (193H12) Rabbit mAb	Cell Signaling Technology	4058S; RRID:AB_331168
Phospho Akt (Thr308) (244F9) Rabbit mAb	Cell Signaling Technology	4056S; RRID:AB_331163
Gapdh (D16H11) XP® Rabbit mAb	Cell Signaling Technology	5174S; RRID:AB_10622025
Rabbit anti-TIAM1 Antibody, Affinity Purified	Bethyl Laboratories	A300-099A; RRID:AB_2271617
Mouse E Cadherin, Unlabeled, Clone 36	bdbiosciences	610181; RRID:AB_397580
C-Peptide (E9C4P) Rabbit mAb	Cell Signaling Technology	51634T
CD133 (Prominin-1) Monoclonal Antibody (13A4), APC, eBioscience™	bdbiosciences	11-1331-81
BD Pharmingen™ FITC Rat Anti-Human CD49f	bdbiosciences	561893; RRID:AB_10894397
Anti-Rac1 antibody, clone 23A8	Millipore Sigma	05-389; RRID:AB_309712
Rabbit Monoclonal anti-DCAMKL1 (Dclk1)	Abcam ab109029	ab109029; RRID:AB_10864128
Chemicals, peptides, and recombinant proteins
CUTANA® High Fidelity 2X PCR Master Mix for CUT&Tag	EpiCypher	15-1018
CUTANA™ pAG-Tn5 for ChIC/CUT&Tag	EpiCypher	15-1017
CUTANA™ E. coli Spike-in DNA	EpiCypher	18-1401
CUTANA® Concanavalin A Conjugated Paramagnetic Beads	EpiCypher	21-1401
Afatinib dimaleate	MedChemExpress	HY-10261
Deoxyribonuclease I from bovine pancreas	Sigma-Aldrich	DN25-100MG
Collagenase from Clostridium	Sigma-Aldrich	C9263-500MG
Collagenase D	Sigma-Aldrich	1.1089E+10
Dispase, neutral protease grade II	Roche	4942078001
Tamoxifen	Sigma-Aldrich	T5648-5G
Trypsin inhibtor from Glycine max (soybean)	Sigma-Aldrich	T9003-250MG
Corn oil	Sigma-Aldrich	C8267-500ML
20% Paraformaldehyde Aqueous Solution, EM Grade	Electron Microscopy Sciences	15713-S
Histo-Clear II	National Diagnostics	HS2021GLL
EGTA (Ethylene glycol-bis(2-aminoethylether)-N,N,N′,N′-tetraacetic acid)	Sigma-Aldrich	E3889-100G
TRIzol LS Reagent	Invitrogen	10296028
Recombinant Human TGFalpha	R&D Systems	239-A-100
Recombinant Mouse Amphiregulin	R&D Systems	989-AR-100
Recombinant Mouse EGF	R&D Systems	2028-EG-200
Recombinant Mouse Epiregulin	R&D Systems	1068-EP-050
ACK lysis buffer	ThermoFisher Scientific	A1049201
Caerulein	Bachem	4030451
Matrigel Matrix	Corning	356231
DAPI (4’,6-Diamidino-2-Phenylindole, Dilactate)	ThermoFisher Scientific	Cat# D3571, RRID:AB_2307445
TDE1 Tagment DNA Enzyme and TD Buffer	Illumina	20034198
Trametinib diet (0.5mg/kg)	Bio-Serv	S10649
PLX4720 (417 mg/kg)	Bio-Serv	S10650
Accumax - Cell Aggregate Dissociation Medium	Invitrogen	00-4666-56
Leibovitz’s L-15 Medium, no phenol red	Gibco	21083027
ImmPRESS HRP Horuse Anti-Rabbit IgG polymer kit Peroxidase	Vector Laboratories	MP-7401-15
ImmPACT DAB substrate kit, peroxidase	Vector Laboratories	ZL0626
Rac1-IN-5	MedChemExpress	HY-175341
XhoI	NEB	R0146L
Critical commercial assays
MinElute PCR Purification Kit	Qiagen	28006
Alcian Blue (pH 2.5) Stain Kit	Vector Laboratories	H-3501
Agilent RNA 6000 Nano Kit	Agilent Technologies	5067-1511
Agilent High Sensitivity DNA kit	Agilent Technologies	5067-4626
iScript™ cDNA Synthesis kit	Bio-Rad Laboratories	1708891
TaqMan Fast kit	Applied Biosystems	4444556
PowerTrack SYBR Green kit	Applied Biosystems	A46109
TaqMan probes	ThermoFisher Scientific	Mm02342486_mH, Mm00465942_m1, Mm00627532_s1, Mm00479622_m1, Hs00413554_m1, Mm00492980_m1, Mm00435565_m1, Mm00448840_m1
Pierce BCA Protein Assay Reagent	ThermoFisher Scientific	23235
Active Ras Pull-Down and Detection Kit	ThermoFisher Scientific	16117
Active Rac1 Pull-Down kit	ThermoFisher Scientific	16118
Mouse Cytokine 32-Plex Discovery Assay®	Eve Technologies	MD32
Mouse Th17 12-Plex Discovery Assay®	Eve Technologies	MDTH17-12
Deposited data
Raw and analyzed data (bulk RNA-seq; ATAC-seq; CUT&TAG)	This paper	GEO:GSE279088/279089
Mouse reference genome assembly MGSCv37 (mm9)	Genome Reference Consortium	http://www.ncbi.nlm.nih.gov/projects/genome/assembly/grc/mouse/
Additional figures (Mendeley Data Figures 1–2)	This paper	Mendeley Data, V1: 10.17632/8fytyyd24t.1
Experimental models: Organisms/strains
Mouse: Mist1CreERT2: Bhlha15 tm3(cre/ERT2)Skz)	The Jackson Laboratory	RRID:IMSR_JAX:029228
Mouse: LSL-KrasG12D: Krastm4Tyj	The Jackson Laboratory	RRID:IMSR_JAX:008179
Mouse: LSL-KrasG12R: Krasem2Ldow	The Jackson Laboratory	RRID:IMSR_JAX:033316
Mouse: LSL-KrasG12V: Krastm1Bbd	The Jackson Laboratory	RRID:IMSR_JAX:026924
Mouse: R26LSL-tdTomato: Gt(ROSA)26Sortm9(CAG-tdTomato)Hze	The Jackson Laboratory	RRID:IMSR_JAX:007909
Mouse: p48-Cre: Ptf1atm1.1(cre)Cvw	The Jackson Laboratory	RRID:MGI:2387812
Mouse: LSL-Trp53R172H: Trp53tm2.1Tyj	The Jackson Laboratory	RRID:IMSR_JAX:008183
Mouse: LSL-Trp53fl: Trp53tm1Brn	The Jackson Laboratory	RRID:IMSR_JAX:008462
Mouse: Rac1P29S: Rac1tm1Jdo	The Jackson Laboratory	RRID:IMSR_JAX:033790
Mouse: Ptf1aTdt: Ptf1atm3.1Mgn/Vu	This paper	N/A
Mouse: R26-myrAkt	Laboratory of Thomas Wunderlich (Max Planck Institute for Metabolism Research)	N/A
Oligonucleotides
Genotyping primer: Ptf1a-F	This paper	CCTTCTGACTTCTCCAAGAAGGCA
Genotyping primer: Ptf1a-R	This paper	CCCTTTATGCCTGGCATTTCACTG
Genotyping primer: Rac1-p29s_CMN-F	Integrated DNA Technologies	CGTGCGGAGTGTTAAATGTC
Genotyping primer: Rac1-p29s_WT-R	Integrated DNA Technologies	TTCTCTCTCAGTCGCTGATTC
Genotyping primer: Rac1-p29s_MUT-R	Integrated DNA Technologies	ACGAGATCAGCAGCCTCTGT
Genotyping primer: AKT-F	Integrated DNA Technologies	TGTCCTAAAGGCCAGCCAAC
Genotyping primer: AKT-R	Integrated DNA Technologies	TTCTCAGTAAGCGTGTGGGC
Genotyping primer: p53-R172H_MUT-F	Integrated DNA Technologies	CCATGGCTTGAGTAAGTCTGCA
Genotyping primer: p53-R172H_CMN-R	Integrated DNA Technologies	GAAACTTTTCACAAGAACCAGATCA
Genotyping primer: p53-R172H_WT-F	Integrated DNA Technologies	AGGTGTGGCTTCTGGCTTC
Software and algorithms
STAR	Dobin et al.63	https://github.com/alexdobin/STAR
Gencode vM1	Harrow et al.64	https://www.gencodegenes.org/mouse/
featureCounts	Liao et al.65	http://subread.sourceforge.net/
FastQC	Wingett et al.66	https://github.com/s-andrews/FastQC
QoRTs	Hartley et al.67	https://hartleys.github.io/QoRTs
Picardtools	Broad Institute	https://broadinstitute.github.io/picard/
limma voom	Law et al.68	https://rdrr.io/bioc/limma/man/voom.html
MarkDuplicates	Broad Institute	http://broadinstitute.github.io/picard/
BEDTools suite	Quinlan laboratory at the University of Utah	http://bedtools.readthedocs.io
MACS2	Zhang et al.70	https://github.com/macs3-project/MACS
DESeq2	Love et al.71	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
HOMER	Heinz et al.72	http://homer.ucsd.edu
pingouin	Vallat73	https://pingouin-stats.org/
fGSEA	Korotkeich et al.74	https://www.biorxiv.org/content/10.1101/060012v3
RPPA SPACE	Shehwana et al.75	https://academic.oup.com/bioinformatics/article/38/22/5131/6751774
Nf-core CutAndRun Pipeline	Ewels et al.79	https://www.biorxiv.org/content/10.1101/610741v3
Bowtie2	Langmead and Salzberg.80	http://bowtie-bio.sourceforge.net/bowtie2/index.shtml
SEACR	Meers et al.81	http://seacr.fredhutch.org/
DiffBind	Ross-Innes et al.82	https://www.nature.com/articles/nature10730
ChIPseeker	Yu et al.83	https://academic.oup.com/bioinformatics/article/31/14/2382/255379?login=true
ImageJ	NIH	https://imagej.nih.gov/ij/
GraphPad Prism 10	GraphPad Software	https://www.graphpad.com/
Adobe Illustrator 2025	Adobe	https://www.adobe.com/
FlowJo Software	BD	https://www.flowjo.com/
Integrative Genomics Viewer	Broad Institute	https://software.broadinstitute.org/software/igv/
Biorender	Biorender	http://biorender.com; RRID: SCR_018361

This paper does not report original code.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR METHODS

Experimental model and study participant details

Animal models.

Mice were housed in a pathogen-free facility at Weill Cornell Medicine (WCM) under standard housing conditions. Mouse lines used were described previously: Mist1-Cre^ERT2 (MGI:3821734 Bhlha15^{tm3(cre/ERT2)Skz})²⁵, LSL-Kras^G12D (MGI:2429948 Kras^tm4Tyj)^55–56, LSL-Kras^G12R (MGI:6281560 Kras^em2Ldow)¹⁸, LSL-Kras^G12V (MGI:3582830 Kras^tm1Bbd)⁵⁷, R26^LSL-tdTomato (MGI:3809523 Gt(ROSA)26Sor^{tm9(CAG-tdTomato)Hze})⁵⁸, p48-Cre (MGI:2387812 Ptf1a^{tm1.1(cre)Cvw})⁵⁹, LSL-Trp53^R172H (MGI:3039264 Trp53^tm2.1Tyj)⁶⁰, LSL-Trp53^fl (MGI:193101; Trp53^tm1Brn)⁶¹. Mice were bred to generate Mist1-Cre^ERT2, LSL-Kras^{G12D or G12R or G12V}, R26^LSL-tdTomato referred to as MK^G12DT, MK^G12RT, MK^G12VT mice, with or without p53^R172H or p53^flox alleles, as indicated. Mice without any of the Kras mutants (G12D, G12R and G12V) are referred to as MT mice. Male and female animals were used for experiments. Ptf1a-Tdt mice (MGI:6470589 Ptf1a^tm3.1Mgn/Vu), generated in this study are deposited at the Vanderbilt Cryopreserved Mouse Repository (https://vcmr.vcscb.org/mouse.html?vger_id=MB). Rac1^P29S mice⁴⁷ were obtained from The Jackson Laboratory (RRID:IMSR_JAX:033790); myrAkt mice⁴⁶ were generously provided by T. Wunderlich (Max Planck Institute for Metabolism Research). Mice developing tumors were treated with buprenorphine as needed; any tumor-bearing mice with discomfort not relieved by pain medication, cachexia, weight loss > 20% initial weight, or distress from any other cause were euthanized in keeping with institutional protocols. All manipulations were performed under the Institutional Animal Care and Use Committee (IACUC)–approved protocol (2017–0038) at Weill Cornell Medicine.

Primary acinar cell culture.

Acinar 3D cultures were generated as described⁶² with few modifications. Briefly, acinar cell isolation was obtained with collagenase/dispase mix dissociation as described below, then cells were filtered through a 100 μm cell strainer.

Method details

Generation of Ptf1a-Tdt mice.

Ptf1a-tdT mice independently expresses tdTomato along with an N-terminal tandem Strep-tag II /Flag tagged version of Ptf1a from the endogenous locus. ES cells containing the Ptf1a Loxed Cassette Acceptor (Ptf1aLCA) allele were used. This allele contains a LoxP site and an inverted LoxP site inside the Ptf1a locus. These sites allow for recombinase mediated cassette exchange. The exchange vector replaced exon 1 of Ptf1a with tdTomato, in frame with a viral 2A peptide, followed by an N-terminal tandem Streptag II/Flag tag fused to the Pft1a cDNA sequence. The Ptf1a^LCA allele was used to insert the Ptf1a^{tdTomato-2A-NSF-ptf1(+HygroR)} cassette using recombinase mediated cassette exchange. Following confirmation of exchange via sequencing, the mice were crossed to a flippase (FLP) recombinase carrying mouse to remove the Hygromycin resistance gene. tdTomato expression is evident in adult acinar cells, but is absent from islets, ducts, and the vasculature. The pancreas appears entirely normal by H&E and mice survive as homozygotes, implying an entirely functional Ptf1a protein. This mouse strain facilitates any experimentation requiring visualization of embryonic pancreatic progenitors cells or adult acinar cells, as well as immunoprecipitation of Ptf1a protein using the FLAG tag. For genotyping, the wildtype allele has a 636 bp band, while the transgenic allele has a 670 bp band. This allows us to distinguish wild-types, heterozygotes and homozygotes. Forward primer: CCTTCTGACTTCTCCAAGAAGGCA; Reverse primer: CCCTTTATGCCTGGCATTTCACTG.

Tamoxifen treatment.

Tamoxifen (Sigma T2859–5G) was dissolved in corn oil and administered by subcutaneous injections (at the indicated ages) at a dosage of 5 mg per injection. Mice were injected once a day for a total of 3 days—administered every other day (total duration: 5 days). Mice were allowed to recover 1 week after the last Tamoxifen treatment before receiving other treatments (if outlined in schema).

Experimental pancreatitis.

For acute injury treatment, mice received 8 hourly intraperitoneal (IP) injections over two consecutive days of either PBS (saline) or caerulein at a dosage of 75 μg/kg diluted in sterile PBS. For the prolonged injury treatment, mice were treated with three hourly IP injections a day, three times a week (on alternating days) for three weeks. For each experiment, either whole pancreata, or a portion of pancreata, was harvested for histologic analysis and downstream -omics assays.

Embryonic pancreatic dissociation.

Wildtype C57BL/6 and Ptf1a-tdT mice were used in all experiments. The fertilization of pregnant females was time-stamped with the discovery of vaginal plugs; Day 0 of embryonic life was assumed to have taken place at noon on the day vaginal plugs were discovered (i.e., days post coitum). Pregnant females were euthanized at the appropriate gestational age with embryos being removed from the uteri, and then placed in a Petri dish containing PBS. Embryos were carefully removed from the uteri with the assistance of a dissection microscope. At E10.5, the excised tissue included pancreatic rudiments (tdT-positive), as well as gut and stomach (tdT-negative). While at E15.5, the pancreas was excised free of surrounding gut and stomach. The excised tissue from both E10.5 and E15.5 were processed into a single-cell suspension using 1 mL of Accumax (Invitrogen, Cat: 00-4666-56) along with approximately 4 μL of 5 mg/mL DNase I [0.02 mg/mL] to minimize cellular clumping. Each tube contained a maximum of 10 embryos to avoid inefficient dissociation. The samples were then incubated in a 37°C water bath for 1 hour, during which the tubes were gently agitated by tapping and trituration to aid in dissociation. After incubation, the tubes were placed on ice and 1 mL of quenching solution was added to each sample (ingredients: supplemented L15 media [see below], DNase I [37.5 μg/mL]) was added to each sample. The tissues were then triturated 7–8 times with a P1000 filtered pipette to achieve thorough dissociation. The resulting cell suspension was filtered through a 35 μm mesh filter into FACS tubes, and the original tube was rinsed with double the volume of diluted quenching solution (ingredients: supplemented L15 media [see below], DNase I [7.5 μg/mL]) to maximize cell recovery. The filtered cell suspensions were centrifuged at 1100 rpm for 10–15 minutes at 4°C, and the supernatant was carefully aspirated, leaving 50–100 μL at the bottom of each tube. Cells were resuspended in quenching solution with DAPI [1:100] for approximately 15–30 minutes before FACS. All samples were processed for FACS analysis immediately after staining to ensure cell viability. Note: the supplemented L15 media used throughout was freshly prepared (ingredients: L15 media w/o phenol red (Gibco, Cat: 21083027), HEPES [10 mM], BSA [1 mg/mL], Pen/Strep [1%]). Similarly, both the concentrated and diluted quenching solutions were prepared fresh.

Mouse pancreatic dissociation.

Mice were euthanized using CO₂, and pancreata were excised and placed in 3 mL of ice-cold HBSS as remaining pancreata were being harvested. Then, the ice-cold HBSS solution was discarded, and pancreata were placed in a 10 cm dish, containing 5 mL of 37°C collagenase and dispase (CD) solution—collagenase D is used for normal tissue; collagenase V is used for fibrotic tissue (CD ingredients: HBSS w/ Ca²⁺ Mg²⁺, Collagenase D / V [1 mg/mL], Dispase II [2 U/mL], STI [0.1 mg/mL], DNase I [0.1 mg/mL]). While in CD solution, pancreata were mechanically dissociated into ~1–3 mm pieces using razor blades. Minced pancreata were then individually transferred to 50 mL conical tubes, and an additional 5 mL of CD solution was added, bringing each sample to a final volume of 10 mL. Tubes were placed on a 37°C orbital shaker at 135 rpm for 25 minutes. After incubation, the samples were centrifuged at 1000 rpm for 3 mins, washed with PBS, and then resuspended and incubated with 2 mL of 0.05% Trypsin-EDTA (37°C) for less than 3 min. Trypsin was inactivated immediately after incubation with 10 mL of 37°C PBS/FBS solution (ingredients: PBS w/o Ca2+ Mg2+, FBS [1:5], STI [0.1 mg/mL], DNase I [0.1 mg/mL]). Each tube was then gently inverted twice, centrifuged, and resuspended in 10 mL of 37°C FACS buffer (ingredients: PBS, EGTA [10mM], FBS [2%], STI [0.1 mg/mL], DNase I [0.1 mg/mL]). Cell suspension was passed through a 100 μm mesh filter into a new 50 mL conical tube, centrifuged, resuspended with 1 mL of FACS buffer solution with DAPI staining [1:100], and then, finally, transferred into a 40 μm FACS filter tube. All samples were kept on ice before and after FACS. For all FACS, Becton-Dickinson Influx or Becton-Dickinson Aria II were used to collect cells of interest. DAPI-/tdTomato+ cells were collected for MT, MK^G12DT, MK^G12RT, and MK^G12VgeoT mice. For ductal cell isolation, cells were stained prior to FACS with Cd49f and Cd133. DAPI-/Cd49f+/Cd133+ cells were collected as ductal cells while DAPI-/CD49f+ cells were collected as tdTomato-negative acinar cell controls.

Bulk RNA-seq.

Following mouse pancreas dissociation, DAPI-/tdTomato+ cells were FACS-sorted directly into TRIzol (100k cells per 750 mL TRIzol) in 1.5 mL microcentrifuge tube. Guanidinium thiocyanate-phenol-chloroform extraction was performed directly and the quality of the samples was determined using the Agilent RNA 6000 Nano Kit on the Agilent Bioanalyzer (Weill Cornell Medicine Genomics Resources Core Facility; WCM-GRCF). The WCM-GRCF prepared libraries using TruSeq RNA Sample Preparation (Non-Stranded and Poly-A selection), and used the NovaSeq 6000 (S1 Flow Cell – Paired End 2×50 cycles) for sequencing. The sequences were aligned to the mouse reference genome (mm9) using STAR4, a universal RNAseq aligner.⁶³ To improve accuracy of the mapping, the genome was created with a splice junction database based on Gencode vM1 annotation.⁶⁴ Sequences that mapped to more than one locus were excluded from downstream analysis, since they cannot be confidently assigned. Uniquely mapped sequences were intersected with composite gene models from Gencode vM1 basic annotation using featureCounts⁶⁵, a tool for assigning sequence reads to genomic features. Composite gene models for each gene consisted of the union of exons of all transcript isoforms of that gene. Uniquely mapped reads that unambiguously overlapped with no more than one Gencode composite gene model were counted for that gene model; the remaining reads were discarded. The counts for each gene model correspond to gene expression values, and were used for subsequent analyses. Prior to the detection of differentially expressed genes, the quality of the sequences was assessed based on several metrics using FastQC⁶⁶ and QoRTs⁶⁷. Differential gene expression analysis was performed for each comparison using limma voom with default parameters.⁶⁸

Bulk ATAC-seq.

Following pancreas cell isolation, 50k DAPI-/tdTomato+ acinar cells were sorted into FACS buffer supplemented with FBS [1:10].⁶⁹ Briefly, acinar cells were centrifuged at 500g for 5 minutes (4°C) and washed with 50 μL of cold PBS. Cells were subsequently lysed using cold lysis buffer, and immediately spun at 500g for 10 min (4°C). The pellet was resuspended in the transposase reaction mix for the Tn5 tagmentation step for 30 minutes at 37°C and sample was purified using a Qiagen MinElute PCR Purification Kit. Next, DNA was indexed and amplified using PCR. The quality of the samples was assessed using Agilent High Sensitivity DNA kit. DNA libraries were then multiplexed and sequenced on a NextSeq2000 (Paired End; P2 – 100 cycles). Data analysis, including quality and adapter filtering, was applied to raw reads using ‘rim_galor’ before aligning to mouse assembly mm9 with bowtie2 using the default parameters. The Picard tool MarkDuplicates (http://broadinstitute.github.io/picard/) was used to remove reads with the same start site and orientation. The BEDTools suite (http://bedtools.readthedocs.io) was used to create read density profiles. Enriched regions were discovered using MACS2⁷⁰ and scored against matched input libraries (fold change > 2 and FDR-adjusted p-value < 0.1). A consensus peak atlas was then created by filtering out blacklisted regions (http://mitra.stanford.edu/kundaje/akundaje/release/blacklists/mm9-mouse/mm9-blacklist.bed.gz) and then merging all peaks within 500 bp. A raw count matrix was computed over this atlas using featureCounts (http://subread.sourceforge.net/) with the `-p` option for fragment counting. The count matrix and all genome browser tracks were normalized to a sequencing depth of ten million mapped fragments. DESeq2⁷¹ was used to classify differential peaks between two conditions using fold change > 2 and FDR-adjusted p-value < 0.1. Peak-gene associations were made using linear genomic distance to the nearest transcription start site with Homer (http://homer.ucsd.edu).⁷²

Acinar 3D cell culture preparation.

Following isolation of acinar cells, cell pellets were resuspended in acinar-explant (AE) media composed of DMEM supplemented with 1% FBS, 1% PS, 1% STI and 1mg/ml of Dexamethasone and then mixed with Matrigel (Corning) in a 2:1 ratio cells:Matrigel. Per 24- well plate, 400 μl of the cells:Matrigel suspension were plated and incubated at 37 °C for solidification for 2–3 hours. Upon Matrigel solidification, 400 μl of warm AE media was added with or without TGF-α at 50ng/ml concentration. Media (with and without TGF-α) was changed at day 1 and day 3 of culture. Brightfield image at 4X and 10X magnification were taken using a Nikon ECLIPSE Ti inverted microscope system equipped with an Andor Zyla 5.5 sCMOS camera.

Analysis of brightfield images of spheroids with computer vision.

Images were acquired in OME-TIFF format with 16-bit depth in grayscale. Illumination was normalized by subtracting the image passed by a gaussian filter of sigma = 5 * max(x_dim, y_dim) / 300, where x/y_dim are the image dimensions. Images were then inverted and intensity values clipped to the intensity between 3 and 99.8 percentiles, scaling image intensity values to unit range. To segment the spheroid objects from the background, we detected edges in the image by Sobel filtering, and seeded the watershed algorithm using values below 0.3 as background and above 0.95 as positive. We dilated objects in this binary mask with a disk of 3 pixels of diameter, and filtered objects smaller than 32 pixels in diameter. We removed further smaller objects by applying closing (dilation followed by erosion), filling holes within objects, and applying erosion and dilation again, all with a 3 pixel diameter disk. All operations were performed using scikit-image version 0.18.2.11 We then quantified various features for each object in each image using the `skimage.measure.regionprops` function, and reduced values per image using the mean. Statistical testing was performed between groups of interest with a two-sided Mann–Whitney U test, and adjusted for multiple comparisons with the Benjamini–Hochberg False Discovery Rate method using pingouin (version 0.4.0).⁷³

Histopathology and immunofluorescence staining.

Pancreata were fixed overnight in 4% buffered PFA, transferred to 70% ethanol, and then embedded in paraffin using IDEXX BioAnalytics laboratory. Serial sections were cut, and then hematoxylin and eosin (H&E) staining was performed. For IF, slides were deparaffinized, underwent an antigen retrieval using Sodium citrate buffer, blocked with 5% BSA supplemented with 0.4% Triton X-100 in PBS, and primary antibodies were incubated overnight at 4°C. Secondary antibodies conjugated with Alexa-488 or Alexa-647 (Invitrogen) were used and DAPI nuclear counterstaining was performed. Fluorescent images were captured with a Nikon ECLIPSE Ti inverted microscope system equipped with an Andor Zyla 5.5 sCMOS camera or entire slide/region area where digitalized with a Zeiss Axioscan 7. For Alcian blue staining, we used the Alcian Blue (ph2.5) Stain Kit (#H-3501) following the manufacturer’s recommendation. H&E and Alcian blue images were captured with a ZEISS Axio Scope.A1 equipped with a Axiocam 105 color or entire slide/region area where digitalized with a Zeiss Axioscan 7.

Gene set enrichment analysis.

Using normalized read counts of RNA-seq data, the fgseaMultilevel function from the fgsea⁷⁴ package with the following parameters were used: minGSSize = 15; maxGSSize = 500 MSigDB Hallmark, MSigDB KEGG, MSigDB REACTOME, and MSigDB Biocarta gene sets were tested for all analyses. Genes were ranked based on DESeq’s wald statistic (stat), which takes into account the log-fold change and its standard error.

Sample preparation and RPPA.

Mice were euthanized using CO₂, and pancreata were excised, washed quickly in 5 mL of ice-cold HBSS to remove excess blood, put into a 1.5 mL microtube and tube was submerged into liquid nitrogen for snap freezing. Briefly, a total of 23 samples, MT, MTK^G12D, MTK^G12R, MTK^G12V mice after saline or caerulein treatment 3 weeks after treatment were processed. Proteins in cellular lysates were denatured using a buffer solution containing 1% SDS and 2-mercaptoethanol, followed by serial dilution in five 2-fold steps with dilution lysis buffer. The diluted lysates were arrayed onto nitrocellulose-coated slides (Grace Bio-Labs) using the Quanterix (Aushon) 2470 Arrayer (Quanterix Corporation), resulting in a total of 5808 spots per slide. These spots included serial dilutions of standard lysates, as well as positive and negative controls prepared from mixed cell lysates and dilution buffer, respectively.

Each slide underwent probing with a validated primary antibody and a biotin-conjugated secondary antibody. The validation process for antibodies used in reverse-phase protein array (RPPA) is detailed on the RPPA Core website (https://www.mdanderson.org/research/research-resources/core-facilities/functional-proteomics-rppa-core/antibody-information-and-protocols.html). Signal amplification was achieved using the Agilent GenPoint staining platform (Agilent Technologies), and the signal was visualized through a DAB colorimetric reaction. Slides were scanned using a Huron TissueScope (Huron Digital Pathology) and quantified with Array-Pro Analyzer software (Media Cybernetics) to obtain spot intensity data.

RPPA data analysis.

Relative protein levels for each sample were determined using RPPA SPACE, a software tool developed by the MD Anderson Department of Bioinformatics and Computational Biology (https://bioinformatics.mdanderson.org/public-software/rppaspace).⁷⁵ This tool fits a logistic model to each dilution curve, utilizing all samples on a slide. The fitting process plots observed and fitted signal intensities against the log2 concentration of proteins for diagnostic purposes. Protein concentrations were then normalized for protein loading by calculating a correction factor through median-centering across samples for all antibody experiments and across antibodies for each sample. Normalization across RPPA sets was performed using replicates-based normalization as described.^76–77

RT-qPCR.

Total RNA was extracted from the sorted cells as specified previously. The concentration and purity of RNA were determined using a NanoDrop spectrophotometer. Reverse transcription was performed using the iScript^™ cDNA Synthesis Kit (Bio-Rad Laboratories, Inc.), and the reaction was carried out in a thermal cycler according to the manufacturer’s instructions. The resulting cDNA was diluted and used as a template for qPCR, which was performed in a 384-well plate format with triplicate wells for each condition. The qPCR was performed using either the TaqMan Fast kit (Thermo) or the PowerTrack SYBR Green kit (Thermo). The following TaqMan primer/probe sets were used: Amy2a3 (Mm02342486_mH), Cpa1 (Mm00465942_m1), Mist1 (Mm00627532_s1), Ptfla (Mm00479622_m1), Hnf6a (Hs00413554_m1), Krt19 (Mm00492980_m1), Pdx1 (Mm00435565_m1), Sox9 (Mm00448840_m1).

Lysate preparation.

Pancreata were excised, washed quickly in 5mL of ice-cold HBSS to remove excess blood and contaminating lymph nodes. Then tissue was put into a 1.5 mL microtube, and then submerged into liquid nitrogen for snap freezing. Frozen tissue was crushed into a powder, and samples were resuspended in 500 μl of RIPA+PIC solution (PIC diluted 1 to 100) and incubated on ice for 15–20 min with intermediate vortex every 5 min. Samples were transferred into cold 1.5 mL tube and centrifuged at 10,000g for 10min at 4°C. Supernatant was transferred to a new tube. Protein concentration quantifications were performed using Pierce BCA Protein Assay Reagent (ThermoFisher Scientific, Cat# 23235) following the supplier protocol.

Western blotting analysis and active Ras/Rac1 pull-down assays.

Standard techniques were employed for immunoblotting. Protein lysates were separated on 4%–12% Bis-Tris NuPage gels (Life Technologies). Western blots were probed with the following antibodies: Anti-Ras Antibody (Active Ras Pull-Down and Detection Kit, ThermoFisher Scientific, 16117), Akt (pan) Rabbit mAb (C67E7, Cell Signaling Technology, 4691S), Phospho Akt (Ser473) Rabbit mAb (193H12, Cell Signaling Technology, 4058S), Anti-beta Actin antibody (Abcam, ab8227), P44/42 MAPK (Erk1/2) Mouse mAb (L34F12, Cell Signaling Technology, 4696S), Phospho-p44/42 MAPK (Erk1/2) Rabbit Monoclonal Antibody (Thr202/Tyr204, D13.14.4E, Cell Signaling Technology, 4370S), Anti-Vinculin antibody (Abcam, ab129002), Anti-Rac1 Antibody (Active Rac1 Pull-Down and Detection Kit, ThermoFisher Scientific, 16118), Vav1 Antibody (Cell Signaling Technology, 2502S), GAPDH Rabbit mAb (D16H11, Cell Signaling Technology, 5174S), and Rabbit anti-TIAM1 Antibody (Bethyl Laboratories, A300–099A). The levels of Ras-GTP were determined using the Active Ras Pull-Down and Detection Kit (ThermoFisher Scientific, 16117) following the manufacturer’s instructions. Similarly, the Active Rac1 Pull-Down and Detection Kit (ThermoFisher Scientific, 16118) was used for determining Rac1-GTP levels according to the manufacturer’s instructions. Detection was performed using chemiluminescence, and images were captured using a Bio-Rad ChemiDoc Imaging System.

CUT&TAG.

Following pancreas cell isolation, one tube of 100k DAPI-/tdTomato+ acinar cells were sorted into FACS buffer supplemented with FBS for each CUT&TAG reaction. CUT&Tag data was carried out following the methods outlined by Henikoff and colleagues.⁷⁸ Data were analysed using the nf-core/cutandrun pipeline v3.0⁷⁹. In brief, reads were aligned to the mouse genome (mm9) using bowtie2 v2.4.4⁸⁰, duplicate reads were removed with Picard v2.27.4, and peak calling was performing using SEACR v1.3⁸¹ with ‘norm’ and ‘stringent’ parameters. Consensus peak sets were created with the R package DiffBind v3.4.11⁸², and differentially modified regions were detected with DESeq2 v1.34.0 (FDR < 0.05). Peaks were annotated with ChIPseeker::annotatePeak v1.30.3⁸³ with respect to Gencode vM1 annotations.

Quantification and statistical analysis

The number of animals is based on feasibility considerations and our interest in observing a relatively large effect size. In addition, a power analysis was performed with a desired power of 0.8, α= 0.05, and a literature search was performed to find expected averages and standard deviations based on similar protocols. Tests for differences between two groups were performed using two-tailed unpaired Student t test or two-tailed Mann–Whitney test as specified in the figure legends. P values were considered significant if less than 0.05. All graphs depict mean ± SEM unless otherwise indicated. Asterisks used to indicate significance correspond with *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. GraphPad Prism 9 or Prism 10 (GraphPad Software) was used for statistical analysis of experiments, data processing, and presentation.

Highlights.

• Inflammation induces lineage reversion in pancreatic acinar cells

• Enhancer rewiring, initiated by injury, is co-opted by Kras^G12D in tumorigenesis

• Kras^G12V/R are not able to sustain enhancer reprogramming and neoplastic commitment

• Deficient Kras^G12R EGFR and RAC1 signaling are rescued by enforced AKT activation